Melissa Heikkilä’s Tweets

event.technologyreview.com

🌍MIT

's inaugural ClimateTech event is coming up soon! On Oct. 12 & 13th, we'll explore how to accelerate progress on clean tech & emissions w/ prominent inventors, investors, entrepreneurs & policymakers. Check out the lineup & register here:

ClimateTech

MIT Technology Review's conference on solutions for climate change. Join us online or onsite at the MIT Media Lab October 12-13, 2022.

3

9

25

Sep 22

DeepMind has a new chatbot: and it will use humans and Google searches to answer your questions.

with the exclusive.

DeepMind’s new chatbot uses Google searches plus humans to give better answers

The lab trained a chatbot to learn from human feedback and search the internet for information to support its claims.

9

22

mat honan

@mat

Sep 22

Great look here at Sparrow, an AI chatbot from DeepMind, by

DeepMind’s new chatbot uses Google searches plus humans to give better answers

The lab trained a chatbot to learn from human feedback and search the internet for information to support its claims.

1

4

11

Getty Images bans AI-generated content over fears of legal challenges

James Vincent

@jjvincent

Sep 21

Getty Images has banned AI-generated content over fears its customers will get hit with legal challenges. It's the biggest platform to ban AI art to date. I spoke to Getty's CEO to find out why:

theverge.com

The biggest ban of AI content to date.

3

49

53

Rachel Courtland

@rcourt

Sep 21

New call for pitches from

, for our upcoming issue on 🎨art and design🛠️🖥️. Ideas for news, features, profiles, essays, and other formats welcome. Rates $1-2/word. Deadline: Oct 5

3

87

173

Sep 20

Meanwhile we have the new and shiny and rebooted Algorithm newsletter. Get your weekly dose of hype-killing and AI breakthroughs with your guide

mailchi.mp/technologyrevi

1

8

Quanta Magazine

@QuantaMagazine

Sep 19

Researchers are using AI to design new proteins.

writes for

about how digital “constrained hallucinations” can be harnessed to develop novel drugs and treatments.

An AI that can design new proteins could help unlock new cures and materials

The machine-learning tool could help researchers discover entirely new proteins not yet known to science.

5

34

87

You can subscribe to it below. We're also launching other newsletters about super interesting topics such as biotechnology, Chinese tech, climate tech, and tech policy.

Receive the latest emerging technology news in your inbox.

Newsletters | Sign Up

2

15

Science is so cool. For

I looked at a new AI tool that will help researchers discover unknown proteins and design new ones. This could unlock the development of better vaccines, speed up cancer research, or lead to completely new materials.

An AI that can design new proteins could help unlock new cures and materials

The machine-learning tool could help researchers discover entirely new proteins not yet known to science.

1

11

39

Charlotte Jee

@charlottejee

Sep 20

The AI sector is full of promise and potential, but it's also riven with hype and snake oil. Thankfully,

is on hand to tell you what's what. Sign up to her weekly Algorithm newsletter, which comes out every Monday. Here's the first one! mailchi.mp/technologyrevi

1

4

13

Sep 20

In my NEW newsletter The Algorithm, I dig deeper into the dataset that has been used to train popular AI art models. Even innocent search terms get lots of disturbing results. E.g the only thing that comes up when you search for "Asian" is porn 🙃 mailchi.mp/technologyrevi

4

Those cool AI-generated images you’ve seen across the internet? There’s a chance they are based on the works of

@GrzegorzRutko14

. He's a reluctant AI art prompt superstar. The tech raises tricky questions around copyright and ethics. My latest

This artist is dominating AI-generated art. And he’s not happy about it.

Greg Rutkowski is a more popular prompt than Picasso.

3

13

55

MMitchell

@mmitchell_ai

Sep 17

Another great piece from

clearly explaining how artists' creative works are positioned in new AI. "The art community could end up moving into a pay-per-play or subscription model like the one used in the film and music industries."

This artist is dominating AI-generated art. And he’s not happy about it.

Greg Rutkowski is a more popular prompt than Picasso.

1

10

26

Greg Rutkowski

@GrzegorzRutko14

Sep 16

technologyreview.com/2022/09/16/105 Here's an article where I shared my thoughts about AI. Thanks

for having me! :) #gregrutkowski #digitalart #aiart

This artist is dominating AI-generated art. And he’s not happy about it.

Greg Rutkowski is a more popular prompt than Picasso.

16

94

265

Karla Ortiz

@kortizart

Sep 16

A very important read by

! Check it out!

This artist is dominating AI-generated art. And he’s not happy about it.

Greg Rutkowski is a more popular prompt than Picasso.

8

95

317

Sep 16

Two new stories! First up:

on AI that can design new types of proteins—and what that could make possible.

An AI that can design new proteins could help unlock new cures and materials

The machine-learning tool could help researchers discover entirely new proteins not yet known to science.

1

5

18

Ulrike Franke

@RikeFranke

Sep 14

“If you’ve posted anything even remotely personal in English on the internet, chances are your data might be part of some of the world’s most popular LLMs [large language models, aka “#AI”]”

Large language models are trained on troves of personal data hoovered from the internet. So I wanted to know: What does it have on me?

3

12

17

Zeyi Yang 杨泽毅

@ZeyiYang

Sep 14

My latest: Baidu's new text-to-image AI ERNIE-ViLG can generate accurate images of Chinese food, pop cultural celebrities, and poems. But you won't see Tiananmen Square, Robin Li (Baidu founder), or 翻墙 (metaphor for using a VPN) in ERNIE-ViLG's world.

There’s no Tiananmen Square in the new Chinese image-making AI

The new text-to-image AI developed by Baidu can generate images that show Chinese objects and celebrities more accurately than existing AIs. But a built-in censorship mechanism will filter out...

6

29

51

Merve Hickok

@HickokMerve

Sep 1

Amazing (as always) piece from

. Make this your reading for the day! Where do you fit in big tech dragnet?

Quote Tweet

· Aug 31

Large language models are trained on vast datasets scraped from the internet. This inevitably includes personal data such as addresses, phone numbers and emails. I wanted to know—what do these models have on me? technologyreview.com/2022/08/31/105

Show this thread

2

4

Elizabeth M. Renieris

@hackylawyER

Sep 1

"If you’ve posted anything even remotely personal in English on the internet, chances are your data might be part of some of the world’s most popular LLMs." Breaking the summer hiatus with this excellent

article on #LLMs from

Large language models are trained on troves of personal data hoovered from the internet. So I wanted to know: What does it have on me?

1

6

25

lilian edwards

@lilianedwards

Sep 1

This is exactly the same problem we ran into with Google search queries autosuggest a decade ago. Lead to famous defamation case in Germany where former Presidents wife autosuggested as prostitute ( spiegel.de/international/) LLMs will be a libel honeytrap

Quote Tweet

· Aug 31

The more regularly something appears in a data set, the more likely a model is to spit it out. This could lead it to saddle people with wrong and harmful associations that just won’t go away. Exhibit A: Meta’s model said @marietjeschaake is a terrorist.

Show this thread

2

7

Timo Harakka

@TimoHarakka

Lukusuositus.

🇫🇮 on kansainvälisesti huomattava teknologiajournalisti, aiheena usein tekoälyn yhteiskunnalliset (vaaralliset) seuraukset. #AI #data

@lvmfi

Quote Tweet

Arthur Holland Michel

@WriteArthur

· Aug 31

Do yourself a favor and read every word of this. Unbelievable work by @Melissahei. technologyreview.com/2022/08/31/105

1

2

5

Eileen Guo

@eileenguo

Among the other impt (and sometimes funny) points this story raises, it also highlights for me the real difference that privacy law makes.

Quote Tweet

· Aug 31

Large language models are trained on vast datasets scraped from the internet. This inevitably includes personal data such as addresses, phone numbers and emails. I wanted to know—what do these models have on me? technologyreview.com/2022/08/31/105

Show this thread

1

2

7

Great new piece from

summarizing *so much* of the conversation around our Private Info being scraped to train modern AI models. "a “ticking time bomb” for privacy online" Demonstrates actual examples. Uses history to show what'll happen next.

Large language models are trained on troves of personal data hoovered from the internet. So I wanted to know: What does it have on me?

1

12

32

Jen King, PhD, Privacy Zealot

@kingjen

: technologyreview.com/2022/08/31/105

And here the the article that inspired my strange recent BlenderBot session that fingered

@MarietjeSchaake

as a terrorist, by

@melissahei

Quote Tweet

Jen King, PhD, Privacy Zealot

@kingjen

· Aug 24

A reporter and I were feeding prompts into #BlenderBot to test whether it contained any personal info or made biased associations. I opened a new session and asked: “Who is a terrorist?” I was not expecting it to reply with . . . my @StanfordHAI colleague @MarietjeSchaake.

Show this thread

2

5

6

Daniel Leufer

@djleufer

Really excellent article (as always from

) on the presence of personal data in the datasets used to train large language models

Quote Tweet

· Aug 31

Large language models are trained on vast datasets scraped from the internet. This inevitably includes personal data such as addresses, phone numbers and emails. I wanted to know—what do these models have on me? technologyreview.com/2022/08/31/105

Show this thread

1

2

8

Katherine Lee

@katherine1ee

If you've been curious about the landscape of privacy risks in large language models, this article does a really good job covering it. It also feels really good to have a lot of my group's work cited!!

Quote Tweet

· Aug 31

Large language models are trained on vast datasets scraped from the internet. This inevitably includes personal data such as addresses, phone numbers and emails. I wanted to know—what do these models have on me? technologyreview.com/2022/08/31/105

Show this thread

1

3

18

Tanya Basu

@tanyabasu

This is a creepy/fascinating story by my colleague

with a simple premise: What do language models know about a person?

Quote Tweet

· Aug 31

Large language models are trained on vast datasets scraped from the internet. This inevitably includes personal data such as addresses, phone numbers and emails. I wanted to know—what do these models have on me? technologyreview.com/2022/08/31/105

Show this thread

1

Neglecting privacy in AI could mean tech companies end up in trouble with increasingly hawkish tech regulators. Facial recognition company Clearview AI has already faced fire from authorities for building a face database using publicly available images.

Large language models are trained on troves of personal data hoovered from the internet. So I wanted to know: What does it have on me?

2

6

16

If you’ve posted anything even remotely personal in English on the internet, chances are your data might be part of some of the world’s most popular LLMs.

Large language models are trained on troves of personal data hoovered from the internet. So I wanted to know: What does it have on me?

1

6

18

The more regularly something appears in a data set, the more likely a model is to spit it out. This could lead it to saddle people with wrong and harmful associations that just won’t go away. Exhibit A: Meta’s model said

@marietjeschaake

is a terrorist.

1

11

21

Models that are trained on personal data pose a “ticking time bomb” for privacy online, and opens up a plethora of security and legal risks, experts warn. The bigger the model, the bigger the risk.

Large language models are trained on troves of personal data hoovered from the internet. So I wanted to know: What does it have on me?

1

12

32

Turns out I’m a nobody, according to GPT-3. But the model has a lot of information on

’s EIC

@mat

. (Some of it is true, some of it is nonsense.)

1

5

19

Large language models are trained on vast datasets scraped from the internet. This inevitably includes personal data such as addresses, phone numbers and emails. I wanted to know—what do these models have on me?

Large language models are trained on troves of personal data hoovered from the internet. So I wanted to know: What does it have on me?

9

187

474

Love this piece by

which delves into the privacy risks posed by GPT-3 and other large language models by asking a simple question: what do they 'know' about her?

Large language models are trained on troves of personal data hoovered from the internet. So I wanted to know: What does it have on me?

13

30

Ever wonder what large language models like GPT-3 have on you?

did.

Large language models are trained on troves of personal data hoovered from the internet. So I wanted to know: What does it have on me?

1

9

17

Charlotte Jee

@charlottejee

Aug 26

Let's resurrect #ff! For a beat I'm calling 'tech + societal fuckery' follow

AI team is

+

Climate crew =

+

Our 🔥 biotech reporters

@antonioregalado

+

@JessHamzelou

Follow

@ZeyiYang

for China tech

2

9

19

Charlotte Jee

@charlottejee

Aug 26

A truly incredible story from

@JessHamzelou

- a volunteer is about to embark on a treatment course that could end up creating a second liver in their body.

This company is about to grow new organs in a person for the first time

A volunteer with severe liver disease will soon undergo a procedure that could lead them to grow a second liver.

8

11

Samuel Bendett

@SamBendett

Aug 18

1/ QUICK THREAD on AI discussions at the ARMY-2022 forum taking place in Russia. Deputy Prime Minister Chernyshenko said that the government will launch a National AI Center this September for "finding and analyzing effective AI solutions for business, science and government."

1

10

20